Human pose estimation for real scenarios. Dealing with multiple people and interaction

DEL PhD Studentship 2013/14

Human pose estimation for real scenarios. Dealing with multiple people and interaction

Principal Supervisor: Jesus Martinez del Rincon

Project Description:

Human pose recovery is one of the most challenging research areas in computer vision. Successful estimations of the human pose provide information and simplify further tasks such as activity recognition or behaviour analysis, and therefore, it can benefit a wide range of industrial sectors such as video surveillance, physical security or sport performance enhancement.

Constant steps forward have been made in this field, solving the problem for a single person performing a simple and repetitive activity in heavily constrained scenarios. However, pose recovery of complex actions in an unconstrained environment still remains a challenging problem. Even more if we consider that interaction
scenarios are of particular interest, e.g. martial arts, boxing, football in the sports domain, fighting, theft, exchanging luggage in the surveillance domain and dancing in the arts domain. To our knowledge, no research has been done for simultaneously recovering the pose of multiple people on video footage.

The aim of this project is to produce a system that tracks the articulated motion of a group of human beings that interact with each other in the scene. The system should be able to work in real life scenarios, such as sport analysis and video surveillance, combining information from multiple cameras. The project will incorporate and combine techniques from different field of expertise, such as activity recognition, action modelling, pose estimation and human tracking.

Objectives:

To develop pose estimation algorithms capable of consider occlusions and interactions between multiple people and its impact in the individual pose estimation, as well as developing strategies to tackle their inherent problems.
To combine activity recognition methods with pose recovery techniques in order to integrate coherently both philosophies, feeding back information to each other and improving the pose estimation accuracy when the activity or the interaction performed by the subjects is identified.
To explore models and algorithms capable of representing interactions between people in the space of activities, by combining non-linear dimensionality reduction methods with advanced Markovian approaches that are able to model interactive activities.
To validate the above approach, by implementing a system that tracks the 3D pose of small groups of interacting people from video footage on a general purpose scenario without pre-assumption of the performed activities.

Academic Requirements:

A minimum 2.1 honours degree or equivalent in Electrical and Electronic Engineering or Computer Science or other relevant subject is required.

General Information

This 3 year PhD studentship, funded by the Department for Employment and Learning (DEL), commences on 1 October 2013, covers approved tuition fees and a maintenance grant (unknown for 2013/14) is approximately £13,000 - £14,000.

Applicants should apply electronically through the Queen's online application portal at: https://dap.qub.ac.uk/portal/

Further information available at: http://www.qub.ac.uk/schools/eeecs/PhD/PostgraduateResearchScholarships/

Contact details:

Supervisor Name:	Jesus Martinez del Rincon
Address:	ECIT, Queen's Road, Queen's island Belfast, BT3 9DT
Email:	j.martinez-del-rincon@qub.ac.uk
Tel:	+44 28(0) 9097 1779
Web:	www.csit.qub.ac.uk